Codex Luna Reserve (gpt-reserve): What It Is, Limits & Troubleshooting
Luna Reserve is extra GPT-5.6 Luna usage that OpenAI gives to selected personal ChatGPT Plus and Pro accounts after regular usage runs out. It uses a separate allowance, does not restore your normal Codex limits, and is not available to every account.
/status on some current Codex clientsOpenAI does not publish a universal Reserve size or fixed reset schedule. For ordinary GPT-5.6 Luna usage, published 5-hour message estimates, and token pricing before limits are reached, see our dedicated GPT-5.6 Luna Codex usage limits guide.
OpenAI's Reserve guidance still identifies GPT-5.6 Luna; verified 2026-09-27. The GPT-6 Sol and Luna rollout does not establish that Reserve has migrated to GPT-6 Luna. The Reserve amount and reset cadence remain unpublished.
Why Did "gpt-reserve" Appear in Your CLI Status Output?
Some current Codex users report seeing Luna Reserve represented as gpt-reserve in /status.
OpenAI officially documents this fallback capability as Luna Reserve. However, public issue reports in the openai/codex repository show that some current CLI clients expose this allocation under the metric name gpt-reserve.
Developers reporting this behavior describe seeing an extra gpt-reserve line alongside their normal 5-hour and weekly meters, showing a remaining percentage and, in some cases, an associated reset indicator. In desktop and web versions of ChatGPT Work, this state is commonly accompanied by a moon icon and a banner indicating that primary plan limits have been reached and that sessions are running on Luna Reserve.
OpenAI's public Help Center article does not document /status or the gpt-reserve label as a guaranteed interface contract. The metric name reflects current client telemetry observed by developers rather than a formal product specification.
Does GPT-Reserve Have a Weekly Limit?
Luna Reserve has its own usage limit, but OpenAI does not document that limit as a guaranteed fixed weekly or seven-day bucket.
OpenAI officially confirms that Luna Reserve operates on a separate allowance with its own usage limit. However:
- OpenAI publishes no universal Reserve token allowance.
- OpenAI publishes no universal Reserve message allowance.
- OpenAI publishes no universal Reserve reset formula.
Search queries frequently refer to a "gpt-reserve weekly limit" because standard Codex subscription allowances operate on weekly reset windows, and some developers observe reset timestamps displayed alongside the reserve meter in their CLI client.
Rather than assuming a universal weekly quota or fixed reset schedule, users should rely on the remaining allowance and any reset timing actually shown in their own account and client interface.
Regular Codex Usage vs. Luna Reserve
Luna Reserve serves as a temporary fallback to keep tasks moving when standard limits are reached. The table below details how regular plan usage compares directly to Luna Reserve:
| Feature / Dimension | Regular Codex Usage | Luna Reserve (gpt-reserve) |
|---|---|---|
| Model Access | Models available under your plan (such as GPT-6 Astra, GPT-5.6 Sol, Terra, and Luna) | Strictly GPT-5.6 Luna; does not provide additional usage for other plan models |
| Trigger Condition | Active while regular plan allowance remains available | Engages as fallback usage after regular usage is exhausted |
| Quota Source | Regular plan allowance, including the 5-hour usage window and applicable weekly limits | Separate fallback allowance (reported in some clients as gpt-reserve) |
| Cost & Billing | Included in your plan subscription | No separate purchase required; included fallback mode for eligible accounts |
| Account Eligibility | Depends on plan or workspace | Selected personal ChatGPT Plus and Pro accounts; availability can change |
| Workspace Support | Personal, Business, and Enterprise workspaces | Personal accounts only; not available in ChatGPT Business or Enterprise |
| Regular Quota Impact | Standard 5-hour recovery and weekly reset schedule | Separate allowance; does not restore regular usage |
| Published Allowances | Official estimated 5-hour local-message ranges for Luna by plan | No universal token, message, or reset allowance published by OpenAI |
(If both your regular quota and Luna Reserve are exhausted, alternative recovery options—such as waiting for when Codex resets, applying a banked reset, or purchasing paid resets and credits—depend on what is available on your specific account.)
Eligibility & Exclusions: Who Gets Luna Reserve?
OpenAI provides Luna Reserve selectively, and access is restricted to personal accounts:
Selected personal accounts on supported Codex and ChatGPT Work experiences. Availability can change and depends on account status and supported app version.
Selected personal Pro accounts. Provides fallback GPT-5.6 Luna usage once regular usage is exhausted. Availability can change and depends on account status and supported app version.
Not available in ChatGPT Business workspaces according to OpenAI documentation.
Not available in ChatGPT Enterprise workspaces according to OpenAI documentation.
Luna Reserve is a fallback feature for selected paid personal subscriptions; not available on free tier accounts.
API usage is billed per token and governed by organization rate limits, completely separate from ChatGPT subscription fallback modes.
Official Availability Guidance: OpenAI documentation notes that availability can change and depends on your account status and supported app version. Reaching a regular usage limit does not automatically guarantee access to Luna Reserve.
How Regular Quota and Luna Reserve Interact
Because regular usage and Luna Reserve operate on separate allowances, consuming reserve capacity follows distinct operational boundaries:
While using Luna Reserve:
- Luna Reserve provides additional usage only with GPT-5.6 Luna.
- It does not provide additional usage for the other models available under your regular plan. Other models may be unavailable while regular usage is exhausted.
- Using Luna Reserve does not restore your regular plan usage.
When regular usage becomes available again:
- When regular usage becomes available again, users can return to the models available under their regular plan. Check your regular Codex usage to see when that happens.
Troubleshooting: Reserve Meter Shows Usage but Codex Won't Run
Current public issue reports in the openai/codex repository show cases where Reserve appears available, yet prompt execution fails. Distinguish below between documented product constraints and observed community reports.
Confirm Luna Reserve Is Actually Active
Luna Reserve provides additional usage only with GPT-5.6 Luna, but merely selecting GPT-5.6 Luna does not establish that Luna Reserve is active or available. After regular usage is exhausted, look for the Luna Reserve notice, moon indicator, or Reserve allowance shown by your supported client and account. If Reserve is active, Luna is the model used for that fallback usage.
Verify that your account displays an active Reserve notice, moon indicator, or reserve meter rather than assuming selecting Luna activates Reserve. Public issue reports show cases where Luna can appear selected while execution is still blocked.
Execution Fails Even When GPT-5.6 Luna Is Selected
Current public issue reports show that Reserve can appear available while requests still fail. The underlying cause is not established by OpenAI's public documentation. A positive reserve percentage shown in client telemetry does not guarantee successful execution.
Low-risk troubleshooting steps include quitting and relaunching your Codex CLI session or desktop application, and verifying that your client is updated to the latest supported release.
Elevated-Command Rejections and Repository Write Failures
Public developer issue reports describe specific failure cases where ordinary sandboxed commands execute successfully while an elevated command is rejected, or where repository inspection and tests succeed while repository writes fail. The underlying cause for these specific execution differences is not documented by OpenAI.
If elevated commands or write operations fail during a session operating under Reserve, run those commands directly in a local terminal outside Codex.
The Reserve Allowance Itself Has Run Out
Luna Reserve has its own usage limit and is not unlimited. If your client shows the Reserve allowance exhausted, Luna Reserve can no longer provide additional Luna usage until capacity becomes available again.
When Reserve is exhausted, you can wait until regular usage becomes available again (when does Codex reset), or check whether alternative options—such as an earned banked reset or paid resets and credits—are available on your account.
Active Turn Continuation vs. Subsequent Prompts
OpenAI officially states that if a usage limit is reached during an active Codex turn, Codex can continue working on that turn subject to fair-use limits. After that active turn completes, users should check the usage page or limit banner for options available on their plan.
If restarting and updating your client do not resolve execution while your client indicates available Reserve capacity, capture your client version, operating system, and a timestamp, and contact OpenAI Support.
Sourcing & Methodology: Official Rules vs. Observed Reports
To maintain strict accuracy on a platform that evolves frequently, this guide separates documented OpenAI policies from observed community findings across their specific first-party sources:
The article "Luna Reserve in Codex and ChatGPT Work" (help.openai.com) supports:
- Selected personal Plus and Pro eligibility
- Exclusion of ChatGPT Business and Enterprise workspaces
- Additional GPT-5.6 Luna usage after regular usage is exhausted
- Separate regular and Reserve allowances
- Luna Reserve having its own usage limit
- Returning to regular-plan models when regular usage becomes available again
The documentation "Using Codex with your ChatGPT plan" supports:
- If a usage limit is reached during an active Codex turn, Codex can continue working on that active turn subject to fair-use limits
Public developer issue reports in the openai/codex GitHub repository support:
- References to the metric label
gpt-reservein client telemetry - Observed
/statuspercentage indicators and reset timestamps - Cases where elevated commands are rejected while sandboxed commands execute
- Cases where repository inspection and tests succeed while repository writes fail
- Instances where execution fails despite positive reserve readings
Frequently Asked Questions
What is Luna Reserve in OpenAI Codex?
Luna Reserve is an automated fallback capacity mode provided by OpenAI for selected personal ChatGPT Plus and Pro accounts in supported Codex and ChatGPT Work experiences. When regular usage is exhausted, Luna Reserve provides additional usage powered strictly by GPT-5.6 Luna so you can continue with Luna while reserve usage remains available. No separate purchase is required. It operates on a separate allowance with its own usage limit, does not restore regular usage, and does not provide additional usage for other models available under your regular plan.
Why did gpt-reserve appear in my Codex /status output?
OpenAI officially documents this feature as Luna Reserve. However, public issue reports in the openai/codex repository show that some current CLI clients expose this fallback capacity under the metric name 'gpt-reserve'. In desktop and web interfaces, this state is commonly signaled by a moon icon and limit notice. OpenAI's public Help Center article does not define gpt-reserve or /status as a guaranteed interface contract.
Does GPT-Reserve have a weekly limit?
Luna Reserve has its own usage limit, but OpenAI does not document that limit as a guaranteed fixed weekly or seven-day bucket. OpenAI publishes no universal Reserve token allowance, message allowance, or reset formula. Search queries frequently ask about weekly limits because regular Codex subscriptions operate on weekly reset windows and some client meters show associated reset timing. Users should rely on the remaining allowance and reset timing actually shown in their own account or client.
Who is eligible for Luna Reserve?
According to official OpenAI Help Center documentation, Luna Reserve is available only to selected personal ChatGPT Plus and Pro accounts on supported Codex and ChatGPT Work experiences. It is explicitly not available in ChatGPT Business or ChatGPT Enterprise workspaces. Availability can change and depends on account status and supported app version.
Does using Luna Reserve restore my regular Codex quota or models?
No. Regular usage and Luna Reserve have separate allowances. Luna Reserve provides additional usage only with GPT-5.6 Luna; it does not restore regular usage or provide additional usage for other models under your plan. When regular usage becomes available again, users can return to the models available under their regular plan.
Why does Codex fail to run even when the Reserve meter shows available quota?
Current public issue reports show that Reserve can appear available while requests still fail. The underlying cause is not established by OpenAI's public documentation. Users should confirm they are using GPT-5.6 Luna, as Reserve provides usage only for Luna, but reports confirm failures can still occur with Luna selected. Low-risk troubleshooting steps include restarting your CLI session or desktop application and ensuring your client is updated to the latest supported release.
Related Guides & Codex Quota Resources
Limits and Models
- Codex usage limits table — Base plan quotas and model allocations across Plus, Pro, and Business.
- GPT-5.6 Luna Codex usage limits — Ordinary Luna usage, published 5-hour message estimates, and token rates.
- GPT-6 Astra Codex limits guide — How frontier reasoning models consume shared plan allowances.
- Codex recommended model and Power presets guide — How to select models explicitly in the CLI and configuration files.
Quota Recovery and Billing
- when does Codex usage reset — Recovery timing for rolling 5-hour and weekly windows.
- Codex banked resets — How promotional reset grants work when available.
- Codex credits, paid instant resets, and API billing — Recovery options when regular quota is exhausted.
- Codex rate limit and 429 triage guide — Error codes, throttling, and cooldown diagnostics.